Studying infantile spasms is challenging because there are so many aspects of variation that introduce potential bias. These might relate to the many underlying etiologies, and variations in clinical semiology and electroencephalographic features that relate more to age or timing of investigation than to the underlying epilepsy or seizures type. New gene defects associated with the CDKL5/STK9 and ARX genes are associated with infantile spasms, but these illustrate that, when studying neurodevelopmental outcomes, it is necessary to deal also with heterogeneity at the level of genotype-phenotype correlation. We discuss these design issues with consideration of data from the United Kingdom infantile spasms study (UKISS)--in which neurodevelopmental outcomes show evidence of an interaction between underlying etiologic classification and randomised treatment--and with consideration to proposals on study design from the recent consensus statement of the West Delphi group. In the continual debate about whether we should "lump" or "split" when studying epilepsy syndromes, we propose the adoption of study designs using valid and consistent methods that permit both lumping and splitting.